Symptoms of Anxiety andSymptoms of DepressionSame Genes, Different Environments?

Kenneth S. Kendler, MD; Andrew C. Heath, DPhil; Nicholas G. Martin, PhD; Lindon J. Eaves, DSc

\s=b\While traditional multivariate statistical methods can de-scribe patterns of psychiatric symptoms, they cannot provideinsight into why certain symptoms tend to co-occur in a

population. However, this can be achieved using recentlydeveloped methods of multivariate genetic analysis. Examin-ing self-report symptoms in a clinically unselected twin sam-

ple (3798 pairs), traditional factor analysis indicates thatsymptoms of depression and anxiety tend to form separatesymptom clusters. Multivariate genetic analysis shows thatgenes act largely in a nonspecific way to influence the overalllevel of psychiatric symptoms. No evidence could be found forgenes that specifically affect symptoms of depression withoutalso strongly influencing symptoms of anxiety. By contrast,the environment seems to have specific effects, ie, certainfeatures of the environment strongly influence symptoms ofanxiety while having little impact on symptoms of depression.These results, which are replicated across sexes, suggest thatthe separable anxiety and depression symptom clusters in thegeneral population are largely the result of environmentalfactors.

{Arch Gen Psychiatry 1987;44:451-457)

Individual psychiatric symptoms are not independentlydistributed in the population. Rather, symptoms tend to

cluster to form recognizable psychiatric syndromes. Al¬though initially the province of the diagnostician, the task ofrecognizing and describing clinical syndromes has beensupplemented, for several decades, by multivariate statis¬tical methods.1,2 These methods can identify syndromes byshowing that certain symptoms often occur together inindividuals in a population; however, they provide no insightinto why these symptoms tend to covary.

In this article, we apply newly developed methods ofmultivariate genetic analysis3 that can move beyond tradi¬tional factor analysis to clarify why certain symptoms tendto cluster. We apply these methods to self-report symptomsof anxiety and depression from a large clinically unselectedtwin sample.4 Our goal is to understand why certain individ¬uals display depressive symptoms, while for others thesymptoms of anxiety are more pronounced.5"13

We wish to test two major hypotheses. The first is thatcertain genes specifically influence the liability to depres¬sive symptoms and other genes specifically influence theliability to symptoms of anxiety. The second hypothesis is

that certain environmental factors are specifically depres-sogenic and others are specifically anxiogenic.

METHODSSample

This study is based on completed postal questionnaires, mailedduring the period from 1980 to 1982, received from 1978 same-sexfemale, and 918 same-sex male, and 902 opposite-sex volunteertwin pairs older than the age of 18 years from the AustralianNational Health and Medical Research Council (NHMRC) TwinRegister, Canberra. As described elsewhere,4 zygosity was deter¬mined by questionnaire items shown to be at least 95% accurate.The questionnaire contained a seven-item anxiety and a seven-itemdepression subscale from the Delusions-Symptoms-States Inven¬tory (DSSI), developed and validated by Bedford et al." Respon¬dents were asked to indicate whether they had experiencedsymptoms "recently": 1, not at all; 2, a little; 3, a lot; and 4,unbearably. The prevalence of symptoms of anxiety and depressionas assessed by this scale was similar in the twin sample and ingeneral population samples from Australia.4 Frequency of contactamong members of a twin pair was shown to be unrelated toconcordance for symptoms. To simplify the analyses, the 902opposite-sex twin pairs were excluded from the multivariategenetic analyses.

Because few individuals checked the most extreme response(unbearably), response categories 3 and 4 were collapsed into a

single category for the purposes of these analyses. Furthermore,because of the low response rate, the last item of the depressionscale (depressed, thoughts of suicide) was eliminated from themultivariate analysis. Since the full text of these items has beenpresented previously,4 in this report, we will use the abbreviateditem versions.

Data Analysis: An OverviewBecause of the statistical complexity of some of the material in

this article, in this section, a relatively nontechnical overview ofthe methods of data analysis is presented. More technical aspectsare outlined in the "Data Analysis: Methods" section. Finally, thefirst paragraph of the "Comment" section contains a nontechnicalsummary of the important results.

There are three major steps to the data analysis presented in thisarticle. First, a traditional factor analysis of the twin responses tothe DSSI items is presented. Second, the fit of various models tothese responses is examined using multivariate genetic analysis.Third, after the determination of the most appropriate multivari¬ate genetic model, the results ofthat model are presented in detail.

Factor analysis attempts to account for the observed correla¬tions between a relatively large number of symptoms in terms ofthe effects of a small number of latent dimensions or factors. Factoranalysis utilizes as "raw" data only the cross-correlations ofsymptoms within individuals. Thus, factor analysis is purely a

descriptive technique that can succinctly summarize patterns ofsymptom covariation. For example, if the DSSI items are providingonly a gross measure of overall "psychiatric distress," we wouldexpect a single-factor solution. If the items are able to discriminatebetween two dimensions of symptomatology (eg, symptoms ofanxiety vs depression), at least two factors would be needed toexplain the observed pattern of symptoms correlations.

The next step in the data analysis is multivariate genetic

Accepted for publication Aug 29, 1986.From the Departments of Psychiatry (Dr Kendler) and Human Genetics

(Drs Kendler, Heath, Martin, and Eaves), Medical College of Virginia,Virginia Commonwealth University, Richmond. Dr Martin is now withQueensland Institute for Medical Research, Herston, Queensland, Aus-tralia.

Reprint requests to Department of Psychiatry, Medical College of Vir-ginia, Virginia Commonwealth University, PO Box 710, Richmond, VA23298 (Dr Kendler).

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

Relationship, as depicted by schematic path diagrams, amonghypothesized genetic factors (G, and G2), hypothesized environ¬mental factors (E, or E, and E2), two hypothesized symptoms ofanxiety (Anx 1 and Anx 2), and two hypothesized symptoms ofdepression (Dep 1 and Dep 2). Strong relationships among varia¬bles are represented by black arrows and weak relationships bygray arrows. In common-pathway model, genetic and environmen¬tal factors affect symptoms by both acting on same latent variable.That is, one genetic (G,) and one environmental (E,) factor specifi¬cally influence latent variable anxiety (Anx), while second genetic(G2) and second environmental (E2) factor specifically influencelatent variable depression (Dep). Individual symptoms are in turninfluenced by latent variables. In this model, genes and environ¬ment, by their influence on latent variables, are equally specific (ornonspecific) in their influence on symptoms of anxiety and depres¬sion. In independent-pathway model, genes and environment di¬rectly and separately influence individual symptoms. One of manypossible configurations is depicted here with this model in whichtwo genetic factors (G, and G2) and one environmental factor (E,)directly influence the four symptoms. G, is relatively specific forsymptoms of anxiety and G2 for symptoms of depression, but E, isnonspecific and influences approximately equally symptoms ofboth anxiety and depression. Thus, in this specific configuration,genes and not environment are responsible for tendency of symp¬toms of anxiety to correlate more highly with other symptoms ofanxiety than with symptoms of depression, and vice versa. Inanother possible configuration of independent-pathway model, andone more consistent with results of this article, environmentalfactors would be relatively specific in their impact on symptoms ofanxiety and symptoms of depression while a genetic factor wouldnonspecifically influence both sets of symptoms.

Com mon-Pathway Model

\/

Anx 1 Anx 2 Dep1 Dep 2

Independent-Pathway Model

Anx 1 Anx 2 Dep1 Dep 2

analysis. This technique can be understood as a generalization offactor analysis that permits the estimation of separate genetic andenvironmental factors. By using information from the correlationsbetween monozygotic (MZ) and dizygotic (DZ) twin pairs for thesame symptom and cross-correlations between and within twinsfor different symptoms, multivariate genetic analysis permits theseparation of the genetic from the environmental impact on

symptom covariation.We wish to test two models in our multivariate genetic analysis

that represent different ways in which genes and environmentmight affect multiple symptoms (Figure). The first, or "common-pathway," model assumes that genes and environment both con¬tribute to one or more intermediate latent variables (eg, liability to"anxiety" and liability to "depression," denoted as "Anx" and"Dep" in the upper section of the Figure), which are in turnresponsible for the observed pattern of symptom covariation. Inother words, this model assumes that genes and environment acton symptom covariation by a final common pathway.

Under the second, or "independent-pathway," model, genes andenvironment may have different effects on the pattern of symptomcovariation. For example (as pictured in the bottom section of theFigure), there could be two sets of genes—one of which wasrelatively selective for symptoms of anxiety and the other forsymptoms of depression—but environmental influences that pre¬dispose equally to both sets of symptoms. It can be shownalgebraically that the common-pathway model can be subsumed asa submodel of the independent-pathway model, so that the fit of thetwo models can be tested statistically (by means of a likelihoodratio 2 test).15

The final step in the "Results" section is to present in detail thefindings of the most appropriate multivariate genetic model. Thispresentation permits a detailed comparison of results between theconventional and multivariate genetic factor analyses and anexamination of the consistency of the findings across sexes.

Data Analysis: MethodsMethods of data summary and analysis designed for continuous

variables are inappropriate for discontinuous variables, such as ouritem scores, which have only three-point scales. The approach thatwe have used assumes the existence, for each item, of a normallydistributed liability that determines the probability of response to

that item. The observed distribution is related to the latentdistribution by abrupt "thresholds" superimposed on the latentdistribution. With multicategory data as those used in this article,it is possible to test statistically the validity of these assumptions.As described previously,4 the fit of this "threshold" model to theobserved data was good.

The first step in our data analysis was a traditional factoranalysis of the twin responses. The sample was subdivided by sexand then into first and second members from each twin pair. Afactor analysis was performed separately for each of the fourresulting subsamples. Factor loadings were estimated by theunweighted least-squares method.14 In each analysis, the numberof factors extracted was determined by the number of eigenvaluesgreater than unity. We estimated uncorrelated ("orthogonal")factors for comparability with the multivariate genetic analysis. Toselect for study one of the infinite number of statistically equiva¬lent solutions ("factor rotations"), we used the simplest techniqueof fixing to 0 the loadings of one depression item ("lost interest ineverything") on the second and third factors, and of an anxiety item("pain or tension in head") on the third factor.15 This method ofrotation ensured comparability of factor rotations between sexes,between first and second twins, and between the traditional andmultivariate genetic factor analyses. These items were chosen byperforming varimax rotations16 on the results from the four sub-samples and then selecting the items for which the mean-squaredfactor loadings were highest on the observed depression andanxiety factors. In fitting three factors, this traditional factoranalysis required the estimation of 36 common factor loadings for13 items on the first latent factor, 12 on the second, and 11 on thethird. Item-specific factor loadings, which explain the variance notaccounted for by the common factors loadings, were obtained bysubtracting from unity the variance accounted for by the commonfactor loadings. By convention, these item-specific loadings are nottabulated.

Although solutions that permit correlated ("oblique") factors aresometimes preferred for descriptive purposes, our chief interestwas in causal analysis for which uncorrelated factors are muchsimpler to interpret. This is particularly true with respect to theaction of different genes that, in the absence of gametic-phasedisequilibrium, should be uncorrelated in the population.

Theoretically, the best data summaries for multivariate analysis

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

Table 1.—Factor Loadings ( 100) of Symptoms of Anxiety and Depressionon Phenotypic Factors in Females and Males*

Females Males

Twin 1 TWin 2 Twin 1 Twin 2

ItemAnxiety subscale

1. Worried about everything 62 37 16 63 28 22 55 37 20 55 33 292. Breathless or heart pounding 37 34 42 42 43 44 35 553. Worked up, can't sit still 57 44 61 34 11 50 50 23 57 37 194. Feelings of panic 66 40 72 27 71 29 63 425. Pain or tension in head 41 44 0t 41 44 0t 49 45 0t 38 59 0t6. Worrying kept me awake 58 31 39 60 25 51 54 25 58 55 27 567. Anxious, can't make up my mind 78 32 79 21 73 16 14 74 34

Depression subscale1. Miserable difficulty with sleep 68 31 74 72 24 63 68 27 52 68 29 572. Depressed without knowing why 70 23 71 18 72 19 72 223. Gone to bed not caring 84 85 83 83 114. Low in spirits, just sat 80 80 10 76 785. Future seems hopeless 83 85 83 18 84 106. Lost interest in everything 89 0t 0t 92 0t 0t 86 0t 0t 91 0t 0t

"Orthogonal factors.tParameter fixed to 0.

of our discontinuous data would be 13-way contingency tables,cross-classifying the scores of individuals on each of the 13 items,for factor analysis, or 26-way tables, cross-classifying responses offirst and second twins on each of the 13 items, for multivariategenetic analysis. In practice, fitting models to such contingencytables, which would require the repeated numerical integration ofthe multivariate normal distribution, would be infeasible withcurrent computer resources. Instead, we have obtained maximumlikelihood estimates of the "polychoric correlation"17 betweenevery pair of variables, separately for each twin group (male andfemale first and second twins for factor analysis; male and femaleMZ and DZ pairs for the multivariate genetic analysis). We thenfitted models to 13 13 or 26 x 26 matrices of polychoric correla¬tions. The factor analyses were performed separately on each13x13 matrix, but the multivariate genetic analysis involvedsimultaneous analysis of two matrices, one for MZ pairs and theother for DZ pairs of a given sex. Models were fitted by unweightedleast squares, in the case of the factor analysis, but by weightedleast squares, using estimates of the reciprocal of the samplingvariance of each polychoric correlation as noniterative weights,18'20for the multivariate genetic analysis. The latter approach givesus an approximate 2 goodness-of-fit test of the absolute fitof the model with the number of degrees of freedom equal to thenumber of unique correlations (650 if we are analyzing two 26 x 26correlation matrices) minus the number of estimated parameters.We can also compute an approximate likelihood ratio 2 (or " 2difference") test of the relative fit of each model compared withmore complete models. For the full model, only a goodness-of-fittest is available. For subsidiary models, the likelihood ratio 2provides a more powerful test. Thus, it is possible that by agoodness-of-fit test a model may provide an acceptable fit to thedata, yet be rejected in favor of a different model by a likelihoodratio test.

In our multivariate genetic analysis using the independent-pathway model, we estimated simultaneously item loadings on thecommon genetic factors, the common (nonfamilial) environmentalfactors, and item-specific genetic factors. Loadings on the commongenetic factors contribute both to the within-individual and to thebetween-twin cross-correlations between items. Loadings on thecommon (nonfamilial) environmental factors contribute to thewithin-individual but not to the between-twin item cross-correla¬tions. Loadings of the item-specific genetic factors contribute tothe correlation between twins for a specific item, but not the cross-correlations between items. Finally, item-specific environmentalfactors, which explain the residual variance, are obtained by

subtraction. Both common and item-specific loadings are expectedto be the same for both members of a twin pair. An independent-pathway model that allows for three common genetic, threecommon environmental, and item-specific genetic factors requiresthe estimation of 85 parameters: 36 (13 +12 +11) common geneticfactor item loadings, 36 common environmental factor loadings,and 13 item-specific genetic factor loadings.

Using the common-pathway model, we estimated as beforecommon genetic, item-specific genetic, and item-specific environ¬mental loadings. However, under this model, the item loadings ofeach common environmental factor are expected to be a constantmultiple of the loadings on the corresponding common geneticfactor. Therefore, it was necessary to estimate only a single scalarmultiplier for each common genetic factor from which loadings onthe corresponding common environmental factor could be derived.In the three-factor common-pathway model, it was thereforenecessary to estimate only 52 parameters: 36 common geneticloadings, three scalar multipliers, and 13 item-specific geneticloadings.

The previous univariate analysis4 indicated that the overall effectof common environmental or genetic dominance on symptoms ofanxiety and depression in this sample was small or undetectable. Ifa variable accounts for a small proportion of variance in an item,statistical principles dictate that it cannot make a major contribu¬tion to the covariation ofthat item with other items. Therefore, ourmultivariate analyses considered only additive genetic and non¬familial (or random) environmental effects, both of which were

shown, in our univariate analysis, to have a large impact on

symptoms of anxiety and depression.4For an estimate of the similarity of factor loadings obtained on

different samples (eg, twin 1 vs twin 2 or males vs females), thecongruency coefficient (rc) was used.21

RESULTSFactor Analysis

Using the eigenvalue criterion, three orthogonal factors wereextracted in each case for the first and second members of the maleand female twin pairs. The results of this traditional, or phe-notypic, factor analysis are seen in Table 1. Factor loadings (which,in an orthogonal solution, are equivalent to the correlation of anitem with the underlying latent factor) are given for the rotatedsolution.

The first phenotypic factor, which accounted for between 45.8%and 50.5% of the total variation, was similar across groups. Thecongruency coefficients were above .99 for all six possible compari-

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

Table 2.—Factor Loadings ( 100) of Symptoms of Anxiety and Depressionon Genetic and Environmental Factors in Female and Male Twins*

Genetic Factors Environmental Factors

Item Specific Specific

Anxiety subscale1. Worried about everything 51

Females

10 33 40 32 21 562. Breathless or heart pounding 31 39 25 29 25 733. Worked up, can't sit still 50 11 35 36 37 594. Feelings of panic 59 14 20 41 26 605. Pain or tension in head 33 34 0t 34 28 34 Ot 686. Worrying kept me awake 40 13 29 29 43 25 39 507. Anxious, can't make up my mind 68 -10 45 25 51

Depression subscale1. Miserable, difficulty with sleep 46 10 45 51 28 50 0*2. Depressed without knowing why 53 13 21 47 21 -16 613. Gone to bed not caring 51 18 13 43 71 174. Low in spirits, just sat 60 17 32 53 -11 465. Future seems hopeless 53 34 69 14 326. Lost interest in everything 63 0t 0t 17 66 0t 0t 37

Anxiety subscale1. Worried about everything 33

Males

46 40 35 632. Breathless or heart pounding 42 44 14 283. Worked up, can't sit still4. Feelings of panic

733874

22 19 36 31 45 13 5815 17 31 55

5. Pain or tension in head 32 34 0t 37 24 44 0t 636. Worrying kept me awake 44 23 29 29 25 72 0t7. Anxious, can't make up my mind 60 13 18 45 25 57

Depression subscale1. Miserable, difficulty with sleep 44 29 48 32 37 502. Depressed without knowing why 57 23 39 35 583. Gone to bed not caring 57 12 56 594. Low in spirits, just sat 65 16 48 10 555. Future seems hopeless 51 -1 32 68 416. Lost interest in everything 62 0t 0t 64 0t 0t 45

"Orthogonal factors, weighted least-square solution.tParameter fixed to 0.^Parameter value constrained to be positive.

sons across the four groups. The highest factor loadings in allgroups were found on four core depression items: "gone to bed notcaring," "low in spirits, just sat," "future seems hopeless," and"lost interest in everything." However, the factor was not highlyspecific for depression as all items loaded positively (ie, > + 0.30)on this factor. This factor was termed "depression-distress" tosignify that depression items consistently loaded highest on thisfactor, but it was also, in part, a general psychiatric distress factor.

The second phenotypic factor, which accounted for between 6.5%and 10.9% of the total variation, was also quite similar in the fourgroups. Five of the six possible congruency coefficients were above.96 and the sixth (between male twin 1 and male twin 2) was .93.The four highest loadings in all groups were from among fiveanxiety items: "worried about everything," "breathless or heartpounding," "worked up, can't sit still," "feelings of panic," and"pain or tension in head." Unlike the first factor, the second factorwas relatively specific. The loadings of all anxiety items except"anxious, can't make up my mind" were in excess of .25, while theloadings for the four core depression items never exceeded. 11. Thisfactor was termed "general anxiety."

A third factor, which accounted for between 5.6% and 5.9% of thetotal variation, had in all four groups by far the highest loading onthe two insomnia items: "worrying kept me awake" and "miserable,difficulty with sleep." Five of the six possible congruency coeffi¬cients were above .90 and the sixth (between female twin 1 and

male twin 1) was .87. This factor was termed "insomnia."A useful way to quantify the contribution of the first two

phenotypic factors to the original anxiety and depression subscalesis to compare the proportion of total variance accounted for in thetwo subscales by the first two factors. Across all four groups, themean ( ± SD) proportion of variance in the anxiety and depressionsubscales accounted for by the "depression-distress" factor was,respectively, 33.8% ±2.8% and 63.4% ±1.9%. In other words, the"depression-distress" factor accounted for one third of the totalvariance of the anxiety subscale, but for nearly two thirds of thetotal variance for the depression subscale. The mean proportion ofvariance in the anxiety and depression subscales accounted for bythe "general anxiety" factor was, respectively, 14.1%±3.0% and2.4% ±0.4%. The "general anxiety" factor accounted for over fivetimes as much variance in the anxiety as in the depressionsubscale.

Multivariate Genetic Analysis: Model FittingWe considered two major multivariate models: the common-

pathway and independent-pathway models (Figure). By a 2 good¬ness-of-fit test, the fit of a "full" independent-pathway model withthree genetic and three environmental factors was excellent forboth females ( 2 = 470.8; d/=565; = .98) and males ( 2 = 556.8;df= 565; = .59). For females, all subsidiary models with fewerthan three genetic and three environmental factors could be

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

rejected by likelihood ratio 2 tests. For males, all subsidiarymodels could also be rejected except that which contained all threeenvironmental factors and only the first two genetic factors(X2 = 16.3;d/=U;P = .13).

The two- and one-factor common-pathway models could berejected at high levels of statistical significance (P<.00001) forboth males and females. However, the three-factor common-path¬way model produced a reasonable fit in both females ( 2 = 550.0;df= 598; = .92) and males ( 2 = 638.4; df= 598; = .12). However,compared with the full independent-pathway model, the three-factor common-pathway model could be rejected by likelihood ratiotests at high levels of significance for both females ( 2 = 79.3;d/=33;P<.0001) and males ( 2 = 81.6; d/=33; P<.0001).

Finally, we fitted the full independent-pathway model to bothsexes simultaneously. The likelihood ratio test of heterogeneitywas very highly significant ( 2 = 222.9; df= 85; P-c.0001), indicatingthat although this model was appropriate for each sex, the factorloadings differed significantly between females and males.

Results of Best-Fitting ModelGenetic and environmental factor loadings are given under the

full independent-pathway model separately for females and formales (Table 2). Although a slightly simpler model also provided anadequate fit in males (ie, two genetic and three environmentalfactors), the full model was somewhat superior in fit and had theadvantage of simplifying the comparison of the results acrosssexes. In comparing these results with the phenotypic factorloadings shown in Table 1, it should be remembered that we are nowfitting a total of six (three genetic and three environmental) factorsrather than three phenotypic factors, so that the individual factorloadings will, in almost all cases, be lower in Table 2 than in Table 1.A comparison of these tables should focus on the pattern ratherthan the absolute value of the factor loadings.

The first genetic factor, which accounted for 26.7% of the totalphenotypic variance in females and 27.3% in males, was verysimilar in both sexes (rc =. 986). The four items with highest loadingin both sexes were two anxiety items, "feelings of panic" and"anxious, can't make up my mind," and two depression items, "lowin spirits, just sat" and "lost interest in everything." Like the firstphenotypic "depression-distress" factor, all items tended to loadhighly and positively on this factor. Unlike the first phenotypicfactor, the average loading for anxiety items was almost as high asthat found for depression items. Because of the apparent lack ofspecificity of this factor, it was termed the "genetic distress" factor.

The second genetic factor accounted for 2.8% of the totalvariance in females and 3.0% in males and was reasonably similaracross sexes (rc= .837). In both sexes, only two items had substan¬tial loadings on this factor: "breathless or heart pounding" and"pain or tension in head." This factor differed from the secondphenotypic "general anxiety" factor in having low loadings forother anxiety items, especially "worried about everything" and"feelings of panic." Therefore, this factor was termed the "geneticsomatic anxiety" factor.

The third genetic factor, which accounted for 2.9% of the totalvariation in females and 3.8% in males, was only modestly stableacross sexes (rc = .510). In females, substantial loadings were seenonly for the two insomnia items. In males, the highest loading wasseen on the first anxiety item "worried about everything, " followedby the two insomnia items. This factor was broadly similar to thethird phenotypic factor and, hence, was termed the "geneticinsomnia" factor. The second and third genetic factors, althoughstatistically significant because of the large size of the sample,account for a small proportion of total variance in liability tosymptoms in the twin population. The genetic specific loadings,which reflect the genetic influences unique to each symptom, were,on the average, relatively modest, accounting for only 7.8% of thetotal variation in liability to symptoms in females and 4.0% inmales. These results suggest that the majority of genetic variancein these symptoms is accounted for by the three extracted factors.

The first environmental factor, which accounted for 24.5% of thetotal phenotypic variance in females and 18.8% in males was similaracross sexes (rc = .984). In both sexes, the four highest loadingswere on the core depression items: "gone to bed not caring," "low inspirits, just sat," "future seemed hopeless," and "lost interest ineverything." This factor was relatively similar to the first phe¬notypic "depression-distress" factor, but the specificity for depres-

sive symptoms was somewhat greater. Therefore, this factor wastermed the "environmental depression" factor.

The second environmental factor, which accounted for 5.8% ofthe phenotypic variance in females and 8.1% in males, was also verysimilar in the two sexes (rc = .986). In both sexes, the three highestloadings were on the core anxiety symptoms "worried abouteverything," "worked up, can't sit still," and "pain or tension inhead." This factor was quite similar to the second phenotypic"general anxiety" factor in loading more equally on all the anxietyitems and hence was termed the "environmental general anxiety"factor.

The third environmental factor, which accounted for 4.0% of thetotal variance in females and 5.3% in males, was also reasonablysimilar in males and females (rc = .835). In both sexes, this factorhad substantial loadings on only the two insomnia items. Thisfactor was broadly similar to both the "insomnia" and "geneticinsomnia" factors and was termed the "environmental insomnia"factor.

For almost all the items, item-specific environmental loadingsthat represent environmental effects (including measurementerror) influencing one item but no others, accounted for a substan¬tial proportion of the total variation. For all items, specificenvironmental variation accounted for 26.0% of the total phe¬notypic variation in females and 30.0% in males.

A useful way to contrast the contribution of the first genetic andenvironmental factors to the anxiety and depression subscales is tocompare the proportion of variance accounted for in these sub-scales by the two factors. The "genetic-distress" factor contributedmore to the total variation in the depression than to the anxietysubscale in both females (29.8% vs 24.1%) and males (31.8% vs23.3%), but the differences were quite small. This is in contrast tothe "environmental depression" factor, which contributed morethan 2V£ times the total variance to the depression than to theanxiety subscale in females (36.3% vs 14.4%). In males, this ratiowas over 3:1 (30.0% vs 9.3%). These results support the conclusionthat the first genetic factor is nonspecific, while the first environ¬mental factor is relatively specific for symptoms of depression.

COMMENTThis article represents, to our knowledge, the first

application of multivariate genetic methods to individualpsychiatric symptoms. We analyzed responses of 3978 twinpairs to the anxiety and depression subscales of the DSSI.Our major goal was to clarify the role of genes vs theenvironment in the etiology of separable anxiety and de¬pression symptom clusters in the general population. Threemajor results are noteworthy. First, a traditional factoranalysis consistently identified two important factorstermed "depression-distress" and "general anxiety." Sec¬ond, in fitting multivariate genetic models, the common-pathway model could be clearly rejected in favor of theindependent-pathway model. Third, fitting the full inde¬pendent-pathway model produced three factors of par¬ticular interest, termed: "genetic distress," "environmentaldepression," and "environmental anxiety." We could findlittle evidence that genes influenced specifically eithersymptoms of depression or symptoms of anxiety. However,certain environments appeared to be specifically depresso-genie and others anxiogenic.

Phenotypic Factor AnalysisIn this large volunteer twin sample, the traditional

eigenvalue criterion readily identified three phenotypicfactors that were stable across four groups (ie, twin 1 and 2in females and males). After rotation, the first of thesephenotypic factors, termed "depression-distress," ac¬counted for about half of the total variation. As the nameimplies, this factor loaded substantially on almost all items,but loadings were consistently highest on the depressionitems. The second phenotypic factor, which accounted forbetween 6% and 11% of the total variance, was termed a

"general anxiety" factor. Loadings for this factor were bothrelatively specific for the anxiety subscale, and were similar

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

for almost all the anxiety items. The third or "insomnia"factor had highest loadings on the two insomnia items withonly quite modest loadings on all other items.

Controversy over the discrimination between symptomsof anxiety and depression has a long history.6,12,22 Two majorviewpoints, which have been termed the "distinct-syn¬drome" and "unitary-syndrome" positions,6 have been ar¬ticulated. The distinct-syndrome position views depressionand anxiety as qualitatively distinct, albeit with some

overlap of symptomatology. The unitary-syndrome view¬point, by contrast, argues that these two states are on a

single continuum, and that any differences between themare basically quantitative and not qualitative. As recentlyreviewed,12,13 empirical studies using a variety of multivari¬ate techniques have tended to support the distinct-syn¬drome position, although these results are not unequivocal.In addition, follow-up studies have strongly supported thediscrimination between anxiety states and depression.10,23

Previous multivariate studies of the relationship betweenanxiety and depression have, with rare exception,24 beenperformed on samples obtained in a treatment setting. Suchan approach introduces an important possible bias. Individ¬uals with symptoms of both disorders are more likely topresent for treatment than those with symptoms from onlyone disorder. This bias can create a spurious covariation ofsymptoms. By contrast, no such bias can be operating in thegeneral population sample studied in this article.

The Australian NHMRC Twin Registry represents a

large, volunteer twin population, in which reported levels ofanxiety and depression do not differ from those observed inthe general Australian population.4 Results from this sam¬

ple provide some support for the "distinct-syndrome" posi¬tion in that two phenotypic factors that could be identifiedas depression and anxiety were extracted from each of thefour subject groups. However, these symptom dimensionswere not completely independent, as anxiety items con¬

sistently loaded positively on the first "depression-distressfactor." By contrast, most depression items had very lowloadings on the second "general anxiety" factor.

Contrary to expectation, consistent evidence was foundfor a third "insomnia" factor. We are unaware of any similarresults that suggest an insomnia factor can be discriminatedfrom anxiety and depression in the general population.These insomnia items, along with other questionnaire dataabout sleep duration and quality, are the focus of anotherreport in preparation.

Multivariate Genetic Model FittingThree aspects of model fitting were examined: (1) the

best-fitting model, (2) the required number of genetic andenvironmental factors, and (3) the consistency of resultsacross sexes. We considered two different models of howgenetic and environmental factors might influence symp¬tom covariation. The first, or common-pathway model,assumed that both genes and environment act on symptomsby influencing the same latent variables. The second, or

independent-pathway model, permitted genes and environ¬ment to influence symptom covariation in different ways.The common-pathway model could be clearly rejected infavor of the independent-pathway model. These findingsindicate that in this sample genes and environment are

influencing the pattern of covariation of individual symp¬toms of anxiety and depression in qualitatively differentways.

The previously reported univariate analysis of thesesymptoms included an examination of the genetic andenvironmental correlation of liability between sexes.4These analyses required the consideration of opposite-sexDZ twin pairs, the inclusion of which in the present multi-

variate analysis would have been extremely cumbersome.In the multivariate genetic analyses, our consideration ofsex differences was limited to showing that, although thesame model produced the best fit in both sexes, the individ¬ual factor loadings differed significantly between the sexes.These results required the separate analysis of results infemales and males, which had the advantage of permittingan assessment of the similarity of results across sexes.

Results of Best-Fitting Multivariate Genetic ModelThe results of the best-fitting multivariate model gave a

striking confirmation of the previous finding that genes andenvironment were influencing symptom covariation in a

qualitatively different fashion. Of the three genetic factors,the first two were relatively stable across sexes, while thethird was only modestly so. The first "genetic-distress"factor was so named because factor loadings were high on allitems with relatively little difference found between de¬pression and anxiety items. Compared with the first phe¬notypic factor, the first genetic factor was substantially lessspecific for depression. This "genetic-distress" factor,which accounted for around 27% of the total phenotypicvariance and over two thirds of the total genetic variance inboth sexes, indicated that genes were largely acting non-

specifically to influence the predisposition to symptoms ofpsychiatric distress.

The second and third genetic factors were quite minor,each accounting for less than 4% of the total phenotypicvariance. The second, or "genetic somatic anxiety" factor,loaded highly on only two anxiety items, both of whichreflected the somatic symptoms of anxiety. This factordiffered from the phenotypic "general anxiety" factor in thelow loadings found for several key symptoms reflectingcognitive aspects of anxiety. Although genes seem to "code"specifically for symptoms of anxiety to a modest degree,they apparently influence only the somatic symptoms ofanxiety.

The third, or "genetic insomnia" factor, was broadlysimilar to the third phenotypic factor in loading mostprominently on the two insomnia items. Genetic factors thatinfluence complaints of insomnia are, at least in part,separable from those that influence general levels ofdistress or symptoms of physical anxiety.

Of the three environmental factors, the first two werestable and the third relatively stable across sexes. The firstor "environmental depression" factor loaded consistentlyhighest on four core depression items. This factor was morespecific for depression than the first phenotypic "depres¬sion-distress" factor, as reflected by the fact that the"environmental depression" factor accounted for over 2%times the total variance in the depression subscale than inthe anxiety subscale.

The second, or "environmental general anxiety" factor,was quite similar to the phenotypic "general anxiety"factor. Loadings were consistently highest on both physicaland cognitive symptoms of anxiety, while loadings were lowon the core depression symptoms. The third, or "environ¬mental insomnia" factor, like the two other insomnia fac¬tors, had highest loadings on the two insomnia items. Theenvironmental factors that influence insomnia also appearto be in part separable from those that cause anxiety anddepression. This is not surprising in that nighttime noisemight be expected to produce precisely this effect.

LimitationsOne potential limitation of this report is noteworthy. The

symptoms studied were obtained by self-report from thegeneral population. As noted above, this has distinct advan¬tages for the kind of multivariate analyses performed. The

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015

use of a population-based sample avoids the possible biasassociated with help-seeking behavior. However, it doesmean that the results obtained here on symptoms ofanxietyand depression cannot necessarily be extrapolated toclinical syndromes. For example, if there were genes spe¬cific for panic disorder, individuals with such genes could berare enough in our sample to prevent detection of a separa¬ble "panic" genetic factor.

SignificanceThe results of this study suggest that the tendency in the

general population for symptoms of anxiety to co-occur withother symptoms of anxiety and symptoms of depression toco-occur with other symptoms of depression is largely theresult of environmental factors. Contrary to our expecta¬tion, genetic influences on these symptoms were largelynonspecific. That is, while genes may "set" the vulnerabilityof an individual to symptoms ofpsychiatric distress, they donot seem to code specifically for symptoms of depression oranxiety. These findings are consistent with a previousanalysis of the total anxiety and depression scale scoresperformed with the Australian NHMRC Twin Registrydata analyzed here.25 In that report, high genetic correla¬tions were found between transformed total scores on theanxiety and depression subscales, indicating that the same

genes were largely responsible for genetic variation in thetwo subscales.

The one notable exception to the apparent nonspecificityof gene action on symptoms of anxiety and depression wasthe consistent emergence of a minor "genetic somaticanxiety" factor. These results suggest that genes may beresponsible for the frequently observed partial indepen¬dence of "somatic" from "psychic" symptoms of anxiety.26

Because measures of relevant environmental variables

were not obtained on twins from the Australian NHMRCTwin Registry, little further information can be extractedfrom the registry regarding the particular environmentalvariables that predispose to symptoms of anxiety vs symp¬toms of depression. However, as indicated by the results ofthe univariate genetic analyses of these data,4 these envi¬ronmental variables were not shared by members of a twinpair. Therefore, the environmental effects that specificallypredispose to symptoms of anxiety vs symptoms of depres¬sion could not plausibly be parental characteristics, towhich both members of a twin pair would be exposed.2'29 Bycontrast, since most life events, except death or illness inrelatives, are not shared by members of an adult twin pair,the results of this study are consistent with findings thatcertain classes of life events specifically precipitate eitherdepression or anxiety.30"32 This study demonstrates thatgenetically informative designs such as MZ and DZ twins,when appropriately analyzed, can not only provide insightinto the role of genetic and environmental factors in theetiology of individual psychiatric symptoms, but can alsoclarify the degree to which the clustering of individualpsychiatric symptoms into syndromes is the result of ge¬netic vs environmental influences.

This study was supported in part by the Department of Mental Health andMental Retardation, Commonwealth of Virginia, and by National Institutesof Health grants AG04954, GM30250, GM32732, HD15838, HL28922,HL31010, and MH40828.

The data on which this report is based were collected with support fromthe National Health and Medical Research Council of Australia who alsosupport the Australian NHMRC Twin Registry. We acknowledge the rolesof J. D. Mathews, MD, PhD, in establishment of the Register and of A. S.Henderson, MD, in collection of the psychiatric symptoms data. We thankRosemary Jardine, PhD, and Marilyn Olsen for substantial help in prepara¬tion of the data.

References1. Fleiss JL, Gurland BJ, Cooper JE: Some contributions to the measure-

ment of psychopathology. Br J Psychiatry 1971;119:647-656.2. Paykel ES: Have multivariate statistics contributed to classification?

Br J Psychiatry 1981;139:357-362.3. Martin NG, Eaves LJ: The genetical analysis of covariance structure.

Heredity 1977;38:79-95.4. Kendler KS, Heath A, Martin NG, Eaves LJ: Symptoms of anxiety and

depression in a volunteer twin population: The etiologic role of genetic andenvironmental factors. Arch Gen Psychiatry 1986;43:213-221.

5. Fahy TJ, Brandon S, Garside RF: Clinical syndromes in a sample ofdepressed patients: A general practice material. Proc R Soc Med 1969;62:331-335.

6. Derogatis LR, Klerman GL, Lipman RS: Anxiety states and depres-sive neurosis: Issues in nosologic discrimination. J Nerv Ment Dis 1972;155:392-403.

7. Derogatis LR, Lipman RS, Covi L, Rickels K: Factorial invariance ofsymptom dimensions in anxious and depressive neuroses. Arch Gen Psychi-atry 1972;27:659-665.

8. Mendels J, Weinstein N, Cohrane C: The relationship between depres-sion and anxiety. Arch Gen Psychiatry 1972;27:649-653.

9. Roth M, Gurney C, Garside RF, Kerr TA: Studies in the classificationof affective disorders: The relationship between anxiety states and depres-sive illnesses\p=m-\I.BrJ Psychiatry 1972;121:147-161.

10. Gurney C, Roth M, Garside RF, Kerr TA, Schapira K: Studies in theclassification of affective disorders: The relationship between anxiety statesand depressive illnesses\p=m-\II.Br J Psychiatry 1972;121:162-166.

11. Prusoff B, Klerman GL: Differentiating depressed from anxiousoutpatients. Arch Gen Psychiatry 1974;30:302-308.

12. Roth M, Mountjoy CQ: The distinction between anxiety states anddepressive disorders, in Paykel ES (ed): Handbook of Affective Disorders.New York, Guilford Press, 1982, pp 70-92.

13. Mullaney JA: The relationship between anxiety and depression: Areview of some principal component analytic studies. J Affect Dis 1984;7:139-148.

14. Bedford A, Foulds GA, Sheffield BF: A new personal disturbancescale (DDSI/SAD). Br J Soc Clin Psychiatry 1976;15:387-394.

15. Joreskog KG, Sobrom D, Lisrel VI: Analysis of Linear StructuralRelationships by the Method of Maximum Likelihood. Chicago, Interna-tional Educational Services, 1983.

16. Harman HH: Factor Analysis, ed 3. Chicago, University of ChicagoPress, 1976.

17. Olsson U: Maximum likelihood estimation of the polychoric correla-tion coefficient. Psychometrika 1979;44:443-460.

18. Heath AC, Berg K: Effects of social policy on the heritability ofeducation achievement, in Berg K (ed): Medical Genetics: Past, Present,Future. New York, Alan R Liss Inc, 1985.

19. Heath AC, Berg K, Eaves LJ, Solaas MH, Sundet J, Nance WE,Corey LA, Magnus P: No decline in assortative mating for education level.Behav Genet 1985;15:349-369.

20. Heath AC, Berg K, Eaves LJ, Solaas MH, Corey LA, Sundet J,Magnus P, Nance WE: Education policy and the heritability of educationalattainment. Nature 1985;314:734-736.

21. Derogatis LR, SerioJC, Cleary PA: An empirical comparison of threeindices of factorial similarity. Psychol Rep 1972;30:791-804.

22. Mapother E: Discussion on manic-depressive psychosis. Br Med J1926;2:872-879.

23. Kerr TA, Roth M, Schapira K, Gurney C: The assessment and predic-tion of outcome in affective disorders. Br J Psychiatry 1972;121:167-174.

24. Murphy JM, Neff RK, Sobol AM, Rice JX, Olivier DC: Computerdiagnosis of depression and anxiety: The Stirling Country Study. PsycholMed 1985;15:99-112.

25. Jardine R, Martin NG, Henderson AS: Genetic covariation betweenneuroticism and the symptoms of anxiety and depression. Genet Epidemiol1984;1:89-107.

26. Schalling D, Cronholm B, Asberg M, Espmark S: Ratings of psychicand somatic anxiety indicants: Interrater reliability and relations to person-ality variables. Acta Psychiatr Scand 1973;49:353-368.

27. Parker G: Reported parental characteristics in relation to traitdepression and anxiety levels in a non-clinical group. Aust NZ J Psychiatry1979;13:260-264.

28. Parker G: Parental characteristics in relation to depressive disorders.Br J Psychiatry 1979;134:138-147.

29. Parker G: Parental representations of patients with anxiety neurosis.Acta Psychiatr Scand 1981;63:33-36.

30. Barrett JE: The relationship of life events to the onset of neuroticdisorders, in Barrett JE (ed): Stress and Mental Disorder. New York,Raven Press, 1979, pp 87-109.

31. Finlay-Jones R, Brown GW: Types of stressful live events and theonset of anxiety and depressive disorders. Psychol Med 1981;11:803-815.

32. Torgersen S: Developmental differentiation of anxiety and affectiveneuroses. Acta Psychiatr Scand 1985;71:304-310.

Downloaded From: http://archpsyc.jamanetwork.com/ by a Universite Laval User on 09/29/2015